The Secret Agent Man

What can Ginkgo do?

Ginkgo is capable of both personal and collaborative (social) learning. Its has the ability to work with individual to:

• learn user's preferences and make recommendations based on this information
• predict user's behavior from a personal history
• learn about a sequence of steps and suggest predict steps to achieve the best result
• learn to classify information and assist in classification decisions
• handle new cases and figure out what to do based on past experience
• give a confidence rating for predictions or conclusions
• learn from other agents
• perform these operations with ambiguous or incomplete input

Ginkgo categorizes cases, or profiles, on a task basis. Unlike simple preference learning, Ginkgo's task-orientation provides the information necessary to improve and optimize workflow processes. This makes it well suited for manufacturing, as-well-as, the typical personal agent applications like shopping.

For example, a Ginkgo-based agent could observe what assembly line workers do under various circumstances to solve problems. When the agent identifies a recurrence of a problem situation, it can either directly fix the problem or recommend the solution to a worker. This can minimize training time, down time, and material waste, all adding up to great savings the manufacturing firm.

This task orientation also lets Ginkgo assist in applications like e-mail management. For instance, when you want to file mail into a folder, an agent could automatically suggest the folder that most closely matches the mail item's subject or contents. The agent could also predict whether incoming mail will require action or not (based on your previous handling) and then sort the items accordingly.

The social learning capabilities of Gingko enable it to:

• promote knowledge sharing
• predict user's behavior from a similar personal history
• find the nearest match for a complex set of inputs (for instance, find the closest user like you)

Ginkgo shares knowledge among users and groups on a situation-by-situation basis. Unlike traditional collaborative filtering, it doesn't assign users to general preference groups using a single statistical technique.  Instead, it uses a technique called "dynamic clustering" to identify similar cases and make recommendations.

Dynamic clustering also allows users to (optionally) determine the source of   recommendations. This is a particularly useful in an applications like the Physician's Consultant. Medical doctors must be able to verify the authenticity and accuracy of suggested treatments. Ginkgo lets the doctor see who made a diagnosis and why.   

Return